African Population Studies Union for African Population Studies ISSN: 0850-5780 Vol. 12, Num. 2, 1997

African Population Studies/Etude de la Population Africaine, Vol. 12, No. 2, September/septembre 1997

Akim J. MTURI

Department of Statistics and Demography, National University of Lesotho, Roma

Code Number: ep97011

ABSTRACT

This paper deals with the determinants of birth intervals in Tanzania, placing emphasis on the effects of the duration of breastfeeding, postpartum amenorrhoea and postpartum abstinence. The analysis is based on the 1991/1992 Tanzania Demographic and Health Survey data. The major determinant of birth intervals in Tanzania is the duration of breastfeeding: the fertility-inhibiting effect of breastfeeding is significant even after controlling for postpartum amenorrhoea and abstinence. It was also noted that the death of the index child creates some additional effects (to those of breastfeeding) on shortening birth intervals. Women who gave birth when they are aged 30 years or more, who reside in urban areas, those in polygamous marriages, those who had a preceding birth interval of at least three years, those working in the modern sector and those whose index child is a boy have longer birth intervals than other women. Furthermore, women residing in West Lake regions have shorter birth intervals, whereas women residing in Southern regions have longer birth intervals than women residing in other regions.

RÉSUMÉ

Ce papier traite des déterminants des intervalles entre les naissances en Tanzanie, en mettant l’accent sur les effets de la durée de l’allaitement au sein, l’aménorrhée après les couches et l’abstinence après l’accouchement. L’analyse se fonde sur les données de l’enquête démographique et de santé de 1991-1992. Le principal déterminant des intervalles entre les naissances en Tanzanie est l’allaitement au sein. L’effet d’inhibition de la fécondité qu’exerce l’allaitement au sein est significatif même si l’aménorrhée et l’abstinence après l’accouchement sont contrôlées. Il a également été découvert que la mort de l’enfant indice (index child) crée des effets additionnels (à ceux de l’allaitement au sein) sur la réduction des intervalles entre les naissances. Les femmes qui mettent un enfant au monde à l’âge de 30 ans ou plus, qui vivent dans les zones urbaines, celles qui sont dans des ménages polygames, celles qui ont eu un intervalle précédent d’au moins trois ans entre les naissances, celles qui sont dans le secteur moderne et celles dont l’enfant indice est un garçon ont un intervalle entre les naissances plus long que les autres femmes. En outre, les femmes qui vivent dans les régions de l’ouest du lac ont un intervalle entre les naissances plus court tandis que celles qui vivent dans les régions sud ont des intervalles entre les naissances plus longs que les femmes vivant dans d’autres régions.

INTRODUCTION

High fertility levels are of major concern to planners and policy makers in most countries in the developing world. Tanzania is no exception. The analysis of birth intervals is of interest in this context since it can provide further insights into the mechanisms underlying fertility change (Njogu and Martin, 1991). Prolonging the intervals between births reduces the number of children a woman can have during her childbearing period (assuming that the age at marriage and the age at which women cease childbearing do not change). Also, short birth intervals raise infant and child mortality ( Hobcraft et al., 1983; Mturi and Curtis, 1995), and this, in turn, can raise the number of births for a woman since infant and child mortality is positively related to fertility in most societies (Preston, 1978).

It is against this background that the Tanzanian government, through the National Population Policy, aims at discouraging pregnancies at intervals of less than two years apart (Planning Commission, 1992). This paper examines the factors associated with birth intervals in Tanzania using the 1991/1992 Tanzania Demographic and Health Survey (TDHS) data. Emphasis is placed on the effect of breastfeeding, postpartum amenorrhoea and postpartum abstinence. The paper suggests possible measures to be considered for prolonging the length of birth intervals.

METHODS AND MATERIALS

The analysis presented in this paper uses data collected in the Tanzania Demographic and Health Survey conducted between October, 1991 and March, 1992. Details of the reproductive history of each woman were collected using the individual women’s questionnaire, together with background information. However, some information (for example, that concerning breastfeeding, amenorrhea and abstinence) was only collected for births since January 1^st, 1986, so it is not possible to include children born before this date in the analysis. Moreover, in order to avoid the problem of the displacement of births from 1986 to 1985 observed in this data set (Mturi, 1996), this analysis includes only intervals which begin with a birth on or after January 1^st, 1987. Since breastfeeding cannot take place before the first birth and postpartum abstinence from sexual intercourse is, by definition, only practised after a birth, the interval between marriage and the first birth has been excluded from the analysis.

The period of interest in this study is that between the date of the birth marking the start of the interval and either the date of the next conception, or the interview date (if no conception occurred before then). If the interval is terminated by a conception, it is termed a closed interval, otherwise it is termed an open interval. In order to avoid any selection bias, open as well as closed intervals are included in the analysis. However, including these intervals will result in pregnancy rates that are biased downwards, because a pregnancy may not be recognized or reported until four or five months after conception (Trussell et al., 1985). In other words, women may have been pregnant at the time of the survey but not been aware of the fact, and hence, not reported to the interviewer. Their birth intervals will be misclassified as open, when they are really closed. The method used to avoid this problem is to artificially ‘’backdate’’ the interview date to nine months before the actual interview date. This study is, therefore, confined to a window of about 4 years: the period from January 1^st, 1987 to nine months prior to the interview date. A total of 6,687 intervals were identified. However, we could not include contraceptive use in the analysis because of a lack of data on exactly when contraception was used. It was necessary, therefore, to eliminate children born to women who have ever used contraception. The final sample (applicable to the population of non-contraceptors) has 4,860 intervals: 2,902 were open and 1,958 were closed.

Since many intervals (all the open intervals) are censored, a hazards model is used for the analysis. There are several types of hazards model available (Kalbfleisch and Prentice, 1980). A Cox proportional hazards model is used here. This choice is supported by the fact that the effect of the majority of covariates on the hazards of conception is proportional for the whole duration since the index birth. Trussell et al. (1985) provide a brief discussion of the application of hazards models to the analysis of birth intervals. The statistical package EGRET (Statistics and Epidemiology Research Corporation, 1990) was used for the analysis.

Covariates

It has been established that breastfeeding has an influence on fertility by lengthening the period of postpartum infecundability (Bongaarts and Potter, 1983) and through its effect on postpartum abstinence (van de Walle and van de Walle, 1993). It has been noted also that the contraceptive effect of breastfeeding is beyond amenorrhea and abstinence. Guz and Hobcraft (1991) have found that a woman who has stopped breastfeeding is more likely to be pregnant once ovulation returns compared with a woman still breastfeeding, due to the reduction of fecundability for breastfeeding women.

Breastfeeding, postpartum amenorrhea and postpartum abstinence are therefore covariates of which their effect on birth intervals change with time. A time-varying structure of eight categories was therefore formed from these three covariates. Women move between the status categories when they finish their periods of abstinence, or breastfeeding, or cease to be amenorrheic. For instance, a woman will move from status 0 to status 2 if she ceases to be amenorrheic whilst still abstaining and breastfeeding, and then from status 2 to status 6 if she resumes sexual intercourse while still breastfeeding. In all cases, a woman starts a birth interval in status 0 and finishes it in status 7, unless the survey date arrives before she reaches status 7. Figure 1 summarizes all possible paths which can be taken by a woman in any interval.

The preliminary analysis showed that only status 6 (breastfeeding only) and status 7 have effects which were significantly different from status 0 (breastfeeding, amenorrheic and abstaining). Therefore statuses 0 to 5 were amalgamated, so that the total number of categories became three: abstaining and/or amenorrheic, breastfeeding only, and none. A woman changes from status 0 to 6 if she stops abstaining or ceases to be amenorrheic while still breastfeeding and finally from status 6 to 7 when she stops breastfeeding. The status of women who stop breastfeeding before amenorrhea and/or abstinence changes directly from 0 to 7 at the point when they cease to be both amenorrheic and abstaining.

One of the reasons why a woman might stop breastfeeding relatively soon after the birth of the index child is if that child dies. It is known that the death of an infant reduces the length of the subsequent birth interval. Clearly, one possible reason for this is that the duration of breastfeeding is reduced. In order to see whether this is the sole reason, we subdivided status 7 (neither breastfeeding, nor amenorrheic, nor abstaining) into two, depending on whether the index child was alive or not. As a result, a time varying structure of four categories was finally used. For convenience, the codes were changed to 3 for women either abstaining or amenorrheic, 2 for women only breastfeeding, 1 for women neither abstaining nor amenorrheic nor breastfeeding with the index child dead, and 0 for women neither abstaining nor amenorrheic with the index child alive.

The fixed covariates used in the analysis, along with their categories, are summarized in Table 1. The administrative regions of Tanzania have been grouped into six geographical zones as follows: Northern (Arusha and Kilimanjaro), Coastal (Dar es Salaam, Coast, Tanga and all regions in Zanzibar), Central (Dodoma, Shinyanga, Singida and Tabora), Southern (Lindi, Morogoro, Mtwara and Ruvuma), West Lake (Kagera, Mwanza and Mara), and Southern Highlands and Western (Iringa, Kigoma, Mbeya and Rukwa). Rural or urban residence is also included as a covariate, since fertility is known to be higher in rural areas than in urban areas (Cohen, 1993). This implies that birth intervals should be shorter for rural women compared with their urban counterparts.

Table 1. Fixed covariates used in analysis

Mother’s education and occupation, and partner’s education were considered as proxy covariates for the socioeconomic status of the household. Usually, women working in the modern sector, highly educated women and/or women with highly educated partners have a high income, better health status, better nutrition and better living standards than less well-educated women. All these factors may influence fertility (Cochrane, 1979). The two education covariates have the following categories: no schooling, lower primary, upper primary, and secondary and above. Occupation depends on how a woman considered her position. Women are classified as ‘’unemployed’’ if they said they were just housewives or claimed to be unemployed. Otherwise, a woman is classified as a ‘’farmer’’ or other ‘’manual worker’’ or an ‘’office worker’’ depending on the occupation she stated.

Other social covariates included in the analysis are religion and type of marriage. It has been shown that Tanzanian women in monogamous marriages have higher fertility than those in polygamous marriages (Mturi and Hinde, 1994). It will be interesting, therefore, to find out if there is a significant difference in the length of birth intervals according to the type of marriage. Also, it has been argued that women belonging to different religious groups may have different actual fertility performances (Bulatao and Lee, 1983): this may be a result of different birth interval lengths.

The sex of the index child (the child whose birth marks the beginning of the interval) is a potential factor determining the birth interval length (Trussell et al., 1985) and is included in this analysis. The length of the preceding birth interval is known to influence the succeeding intervals. That is, long preceding birth intervals are associated with long succeeding birth intervals (Trussell et al., 1985). In this study, preceding birth intervals are grouped into four categories: no preceding birth, less than 24 months, 24-35 months, and 36 months and above. Finally, the mother’s age at the birth of the index child and the order of the index child are hypothesized to be associated with birth interval lengths. These two factors have potentially different effects on birth intervals but, in practice, they are highly correlated (particularly in societies with high fertility and the near universal and early marriage of women). Therefore, it is important to explore both in the analysis, although it is expected that only one will appear in the final model. Four categories of maternal age were created (<20 years, 20-29, 30-34 and 35 above) and birth order is classified into five categories (1, 2, 3-4, 5-6 and 7 and above).

RESULTS

In a proportional hazards model, the covariates are assumed to act multiplicatively on the baseline hazard and the effect is constant over time. Table 2 gives the relative risks along with their corresponding 95 per cent confidence intervals for a parsimonious model containing the covariates which were significant at the 5 per cent level.

To understand the effect of breastfeeding, abstinence and amenorrhea on the hazard of conceiving, we define the reference category to consist of women who are neither breastfeeding, nor abstaining, nor amenorrheic and whose index child is still alive. Compared with women in the reference category, the risk of conception is 73 per cent lower for women who are abstaining and/or amenorrheic (whether or not they are breastfeeding), and 53 per cent lower for women who are breastfeeding, but neither amenorrheic nor abstaining. In other words, breastfeeding alone reduces the risk of getting pregnant by more than half. These results suggest that the fertility-inhibiting effect of breastfeeding is still high even after controlling for postpartum amenorrhea and postpartum abstinence.

Table 2: Estimated relative risks and 95 per cent confidence interval for parsimonious hazards models

Women who ceased breastfeeding and child died have 128 per cent higher risk of conception than women who ceased breastfeeding and child is alive. It might be pointed out that the children of women who do not breastfeed at all (or for a very short duration) are more likely to die, a finding which disturbs the implied causality in the argument here. We know from our data, however, that 98 per cent of Tanzanian women breastfeed their children for sometime, and well over 90 per cent breastfeed for at least 12 months (provided that the child survives for that period of time). This suggests that it is the death of a child, rather than a woman’s voluntary decision not to breastfeed, which is usually the first event in the chain of events leading to short birth intervals. Therefore, the death of a child has an independent effect on the risk of getting pregnant.

The risk of conception is 18 per cent lower for women residing in urban areas compared with their rural counterparts. Zone of residence also has a significant effect on birth interval lengths. Whereas women residing in the Southern zone have a lower risk of conception, women residing in the West Lake zone have a higher risk of conception than women residing in the Central zone. Women residing in the Northern, Coastal, Southern Highlands and Western zones do not have a significantly different risk of conception than women in the Central zone. These results support the findings on regional fertility differentials which suggest that West Lake regions have higher fertility and Southern regions have lower fertility than most of the other regions (Mturi and Hinde, 1995).

Maternal age is found to have a significant effect on birth interval lengths. The risk of conception is 23 per cent lower if a woman’s age at the birth of the index child is between 30 and 34 years than it is if a woman’s age is between age 20 and 29 years, and it becomes 49 per cent lower if a woman’s age at the birth of the index child is 35 years or over. The risk of conception is not statistically different for women who started an interval while aged under 20 years compared with those who started while aged 20-29 years. These results are not surprising: they merely confirm that fecundity of a woman declines after the age of 30 years.

The risk of conception for women in polygamous marriages is 13 per cent lower than women in monogamous marriages. This suggests that overall, women in monogamous marriages have shorter intervals than women in polygamous marriages. This is to be expected since women in polygamous marriages are likely to have a reduced frequency of intercourse compared with women in monogamous marriages. Also, women in polygamous marriages are more likely to adhere to other traditional customs and values (that tend to prolong intervals between births) which are not controlled in this analysis (such as other types of abstinence).

Women working in the modern sector have 56 per cent lower risk of conception than farmers. But, women reported to be unemployed or with manual jobs do not have a statistically different risk of conception compared with farmers. Although this finding is expected, it should be interpreted with caution because the number of cases for women working in the modern sector is very small. The analysis has shown that the highest level of education for both the mother and her partner has weak effects on birth interval lengths. This implies that Tanzanian women have minimal difference in child-spacing practices according to their or their partner’s level of education. It should be noted, however, that highly educated women or women with highly educated partners are more likely to use contraception and hence be excluded from our sample.

As expected, women with preceding birth intervals of 36 months or above have 18 per cent lower risk of conception than women with preceding intervals of 24 to 35 months. Furthermore, intervals which begin with a girl are more likely to be shorter than intervals which begin with a boy. Finally, both birth order of the index child and religion of the mother do not have a significant effect on birth intervals among Tanzanian women included in the sample.

CONCLUDING REMARKS

This analysis of the durations from birth to the next conception has shown that breastfeeding is one of the major determinants of birth interval lengths in Tanzania. The fertility-inhibiting effect of breastfeeding is significant even after amenorrhea ceases and after resuming sexual relations. The results further suggest that mothers whose index child has died have shorter birth intervals than mothers whose index child is alive. Maternal age, occupation, and type of marriage of the mother as well as of the child and previous birth intervals have been found to affect the length of birth intervals. Women who give birth at ages 30 or over, who are in polygamous marriages, and who work in the modern sector have been found to have longer birth intervals than other women. Furthermore, women with previous intervals of at least three years and those with intervals which begin with a boy have longer birth intervals than other women. It has been found also that the place where a woman resides influences the length of intervals between her births. As expected, women residing in urban areas have longer birth intervals than their rural counterparts. Furthermore, women residing in Southern regions have longer birth intervals, whereas those residing in West Lake have shorter birth intervals than women residing in other parts of the country.

The analysis has also demonstrated that any policy aimed at lengthening birth intervals in Tanzania should encourage breastfeeding. Currently, the median duration of breastfeeding in Tanzania is about 22 months. Unfortunately, we could not include contraceptive use in the analysis, because of a lack of data on exactly when contraception was used. However, this is not likely to change the results for traditional child spacing practices very much since Tanzania has a very low contraceptive prevalence. A similar study has included all women (whether they have been contracepting or not) and found results which are very similar to those found in this study (Mturi, 1996).

It is clear from these results that a reduction in the typical duration of breastfeeding in Tanzania will, other things remaining equal, lead to shorter birth intervals. If such a reduction should occur, it will mean that the task of achieving a decline in fertility is made more difficult, since a portion of the increased contraceptive prevalence rates which are a goal of the Tanzania Family Planning Policy will simply be compensating for the reduction in breastfeeding, rather than contributing to a reduction in fertility. It should be noted, however, that the birth intervals are fairly long in Tanzania. According to the 1991/92 TDHS, the median number of months since previous birth for second order births and above was 33.3 months.

REFERENCE

• Bongaarts, J. and R.G. Potter. 1983. Fertility, Biology and Behaviour: An Analysis of the Proximate Determinants. Academic Press, New York.
• Bulatao, R.A. and R.D. Lee. 1983. ‘’A Framework for the Study of Fertility Determinants’’. In R.A. Bulatao and R.D. Lee (eds) Determinants of Fertility in Developing Countries: Supply and Demand for Children, volume 1:1-26. Academic Press, New York.
• Cochrane, S.H. 1979. ‘’Fertility and Education: What Do We Really Know ?’’ World Bank Staff Occasional Papers, number 26.
• Cohen, B. 1993. ‘’Fertility Levels, Differentials, and Trends.’’ In K.A. Foote, K.H. Hill and L.G. Martin (eds) Demographic Change in Sub-Saharan Africa, p:8-67. National Academy Press, Washington, D.C.
• Guz, D. and J. Hobcraft. 1991. ‘’Breastfeeding and Fertility: A Comparative Analysis.’’ Population Studies, 45:91-108.
• Hobcraft, J. ; J.W. McDonald and S. Rutstein. 1983. ‘’Child-Spacing Effects on Infant and Early Child Mortality.’’ Population Index, 49:585-618.
• Kalbfleisch, J.D. and R.L. Prentice. 1980. The Statistical Analysis of Failure Time Data. John Wiley, New York.
• Mturi, A.J. 1996. ‘’Patterns of Fertility and Contraceptive Use in Tanzania’’. Unpublished Doctoral Thesis, Southampton, U.K.: University of Southampton.
• Mturi, A.J. and S.L. Curtis. 1995. ‘’The Determinants of Infant and Child Mortality in Tanzania.’’ Health Policy and Planning, 10:384-394.
• Mturi, A.J. and P.R.A. Hinde. 1994. ‘’Fertility Decline in Tanzania. ‘’ Journal of Biosocial Sciences, 26:529-538.
• Mturi, A.J. and P.R.A. Hinde. 1995. ‘’Recent Demographic Change in Tanzania: Causes, Consequences and Future Prospects.’’ Journal of International Development, 7:117-134.
• Njogu, W. and T.C. Martin. 1991. ‘’Fertility Decline in Kenya: The Role of Timing and Spacing Births.’’ Proceedings of the Demographic and Health Surveys World Conference, August, 5-7, Washington D.C. Proceedings, Volume III:1883-1901.
• Planning Commission. 1992. National Population Policy. President’s Office, The United Republic of Tanzania.
• Preston, S.H. 1978. ‘’Introduction.’’ In S.H. Preston (ed.). The Effects of Infant and Child Mortality in Fertility. Academic Press, New York, p:1-20.
• Statistics and Epidemiology Research Corporation. 1990. EGRET Reference Manual.
• Trussell, J.; L.G. Martin; R. Feldman; J.A. Palmore; M. Conception and D.N.A. Bakar. 1985. ‘’Determinants of Birth-Interval Length in the Philippines, Malaysia, and Indonesia: A Hazard-Model Analysis.’’ Demography, 22:145-168.
• Van de Walle, E. and F. Van de Walle. 1993. ‘’Post-Partum Sexual Abstinence in Tropical Africa.’’ In R. Gray; H. Leridon and A. Spira (eds.). Biomedical and Demographic Determinants of Reproduction, Clarendon Press, Oxford, pp:446-460.

Copyright 1997 - Union for African Population Studies.